You are on page 1of 46

RESEARCH ON SYSTEM ZEROS: A SURVEY∗

Cheryl B. Schrader and Michael K. Sain


Department of Electrical and Computer Engineering
University of Notre Dame
Notre Dame, Indiana 46556

ABSTRACT

This report provides a review of notable publications regarding the topic of zeros of
multivariable systems in the period from 1970 into 1989. Particular attention is paid to
relationships between definitions of such zeros and their theoretical development. Trans-
mission, invariant, system, module, infinite, and finite zeros are addressed, the latter subdi-
vided into a hierarchical structure with respect to information considered in the definition
process.


International Journal of Control, Volume 50, Number 4, Pages 1407-1433, 1989. This work was sup-
ported in part by the University of Notre Dame’s Frank M. Freimann Chair in Electrical Engineering.
The work of the first author was also supported by the SAE Forgivable Loan Program; that of the second
author also received support from The Ohio State University’s Distinguished Visiting Professorship.
I. INTRODUCTION

During the past two decades, the concept of zeros of multivariable systems has received
considerable attention from researchers. Myriads of multivariable zeros have been defined
in the literature and, as a result, the term zero has become ambiguous. In this light, this
survey provides a historical development of the theory pertaining to zeros of multivariable
systems.

Although the first decade of investigation into this topic provided some comparisons
regarding zeros development, a comprehensive survey of various zero types and approaches
is not available to the knowledge of these authors. In 1975 B.A. Francis and W.M. Wonham
provided an appendix describing some definitions of zeros of linear multivariable systems
in order to support the choice of definition upon which their research is based. For similar
reasons, A.G.J. MacFarlane and N. Karcanias (1976) published a survey of the algebraic,
geometric and complex-variable theory of poles and zeros of linear multivariable systems.
Finally, in 1977, A.C. Pugh discussed some definitions of multivariable zeros. In all of
these papers, only a class of zeros related to the problem at hand was considered; for
example, Pugh excluded certain definitions of zeros confined to the state space point of
view. In addition, many developments in the investigation of multivariable zeros have
occurred during the past decade.

The importance of the study of zeros of multivariable systems was in evidence early in
control theory history (Athanassiades et al. 1962, Davison 1969). Performance tradeoffs
in the design of feedback control systems containing zeros positioned in the dominant
time-frequency frame have been accepted conventionally. In order to achieve a stable
feedback system, certain bounds or limitations on response are imposed by those zeros
which lie in the right-half complex plane (Cheng and Desoer 1980, Bristol 1981, Sain et
al. 1981a,b, Antsaklis and Sain 1981, Kimura 1982, Clarke 1984, Freudenberg and Looze
1985, Postlethwaite and Foo 1985, Vidyasagar 1986).

This paper characterizes zeros of linear multivariable systems by establishing a hier-


archy of information involved in the definition process. In order to define multivariable

2
zeros, how much of the entire structure is used? Are the zeros defined merely as complex
numbers? Do the definitions use invariant factors or module theory, address multiplicity or
zero directions? Advantages and difficulties of these various approaches are presented and
discussed.

In order to establish a basis for discussion, the rudimentary algebraic idea of vector
spaces over fields as well as a familiarity with multivariable problems in the frequency
domain is necessary. General systems satisfying the state equations

ẋ(t) = Ax(t) + Bu(t) (1)


y(t) = Cx(t) + Du(t)

with x(t)  Rn , u(t)  Rm , y(t)  Rp are of interest. The (Rosenbrock) system matrix
corresponding to (1) is " #
X sI − A B
= . (2)
−C D
Expression (2) includes the system’s internal structure and conveys more information than
the multivariable transfer function matrix.

Assuming the initial conditions on the state vector x in (1) are zero and taking the
Laplace transform of both members of (1), we obtain

ŷ(s) = [C(sI − A)−1 B + D]û(s). (3)

If we rewrite (3) as
ŷ(s) = G(s)û(s),

we have obtained an external description of the system (1) where G(s) : U(s) → Y (s)
and û(s)  U(s), ŷ(s)  Y (s) with U(s) and Y (s) vector-spaces over R(s), the field of
polynomic ratios in s with real coefficients. More technically, if U and Y are R-vector
spaces of dimension m and p, respectively, the sets U(s) (and Y (s)) can be viewed as
tensor products R(s) ⊗ U (and R(s) ⊗ Y ) of R-vector spaces in the customary manner. By
construction, we know that G(s) is a p × m matrix of proper rational transfer functions
with real coefficients and having reduced representatives.

3
Similarly, discrete systems of the form

xi+1 = Axi + Bui


yi = Cxi + Dui

with xi  Rn , ui  Rm , yi  Rp can be associated with both a system matrix and a transfer


function matrix G(z) using z-transforms, assuming zero initial conditions. The discrete
system can also be identified by the matrices A, B, C, and D. For ease of notation, s is
written as the transformed variable and should be interpreted in context.

The transfer function matrix G(s) can be obtained directly from the system matrix
by rational operations over R(s) (Rosenbrock 1970). The following series of operations
produces the standard form for G(s):
" # " #
X sI − A B I (sI − A)−1 B
= −→
−C D −C D
" #
I (sI − A)−1 B
−→
0 C(sI − A)−1 B + D
" #
I (sI − A)−1 B
−→
0 G(s)
" #
I 0
−→ .
0 G(s)

In a similar manner, these system representations are a basis for the module theoretic
approach to systems. Not only are the Rosenbrock system matrix and its resulting transfer
function matrix an integral part of this investigation, operations over k[s], the ring of poly-
nomials in s over k, transform the matrices into desirable form. The dynamical behavior
of the system is then evaluated in terms of modules over k[s] and the more familiar vector
spaces over k(s), where k is an arbitrary field.

The topic of zeros of linear multivariable systems has prompted considerable interest in
the recent literature. The study of poles and zeros in the single-input, single-output (SISO)
case has been exhaustive. Consider a SISO transfer function
n(s)
g(s) = .
d(s)

4
Then the zeros of the relatively prime polynomials n(s) and d(s) are called the zeros and
poles of the transfer function, respectively. However, when one considers the multivariable
case, many different approaches to the definition of zeros exist.

One common extension of the SISO case is to consider each element of a transfer func-
tion matrix G(s) to be an individual transfer function between a certain input and a certain
output. Thus SISO theory can be applied easily. In some investigations, the multivariable
system zeros have been defined in terms of individual element zeros (Davison 1970). How-
ever, this is not often the case. Many zero generalizations are based upon a representation
of a system or transfer function matrix; two standard canonical forms, the Smith form and
the Smith-McMillan form are theoretical tools used in some definitions of multivariable
zeros.

For any p × m polynomial matrix P (s) with rank r, pre and post multiplication by
appropriate unimodular polynomial matrices U1 (s) and U2 (s) as in

U1 (s)P (s)U2 (s)

produces an equivalent matrix Λ(s) of the form


" #
Λ∗ (s)r,r Or,m−r
Λ(s) = . (4)
Op−r,r Op−r,m−r
The subscripts in (4) denote submatrix dimensions and

Λ∗ (s) = diag{λ1 (s), λ2 (s), ..., λr (s)}. (5)

Each λi (s) in (5) is a unique monic polynomial satisfying

λi (s) | λi+1 (s); 1 ≤ i ≤ r − 1

where a | b indicates that there exists a polynomial c such that b = ac. If we define ∆i (s)
to be the monic greatest common divisor of all the i × i nonzero minors of P (s), then

λi (s) = ∆i (s)/∆i−1 (s); ∆0 (s) = 1.

The matrix Λ(s) is identified as the Smith form of P (s), the {∆i (s)} are referred to as
the determinantal divisors of P (s), and the {λi(s)} are the invariant polynomials of P (s)
(Kailath 1980).

5
A canonical form, directly resulting from the existence of the Smith form, for matrices
whose elements are rational functions is the Smith-McMillan form. A p×m transfer function
matrix G(s) of rank r can be rewritten as

N(s)
G(s) =
d(s)

where N(s) is a polynomial matrix of rank r and d(s) is the least common monic denomi-
nator of the elements of G(s). We can then write

N(s) = U1 (s)Λ(s)U2 (s)

where U1 (s) and U2 (s) are again unimodular matrices and Λ(s) is in Smith form. The
matrix M(s), called the Smith-McMillan form of G(s), is characterized by

Λ(s)
M(s) = U1−1 (s)G(s)U2−1 (s) = .
d(s)

Reducing the elements of M(s) produces

λi (s) εi (s)
=
d(s) ψi (s)

with {εi (s), ψi (s)}, 1 ≤ i ≤ r, coprime. M(s) is of the form

" #
M ∗ (s)r,r Or,m−r
M(s) =
Op−r,r Op−r,m−r

with
( )
∗ εi (s)
M (s) = diag ; 1 ≤ i ≤ r.
ψi (s)

In the Smith-McMillan form we find the following three properties:

(i) εi (s) | εi+1 (s), 1 ≤ i ≤ r − 1


(ii) ψi+1 (s) | ψi (s), 1 ≤ i ≤ r − 1
(iii) d(s) = ψ1 (s).

6
II. CLASSIFICATION AND MOTIVATION

Although many authors before 1970 alluded to the concept of zeros of multivariable
systems (see, for example (Brockett 1965, Simon and Mitter 1969)), H.H. Rosenbrock is
credited with the first definition of zeros of a multivariable system or, loosely termed,
multivariable zeros. The original definitions of Rosenbrock will be examined in great detail
as they provide a means of classification of multivariable zeros.

Rosenbrock’s first multivariable zeros were distinctly related to the standard state space
tests for controllability and observability (Rosenbrock 1970). These zeros were termed
decoupling zeros. The complex numbers {s0 } which satisfy

rank[s0 I − A B] < n (6)

are called input decoupling zeros (i.d. zeros). Similarly, those complex numbers {s0 } which
satisfy

" #
s0 I − A
rank <n (7)
−C

are called output decoupling zeros (o.d. zeros). The number of i.d. zeros is equal to the
row rank defect of the controllability matrix

[B AB A2 B · · · An−1 B];

and the number of o.d. zeros is equal to the column rank defect of the observability matrix
 
C
 
 CA 
 
 . 
 .
 . 
 
 
 . 
CAn−1

P
Consider the system matrix 1 obtained by removing i.d. zeros from a given system
P
matrix by the order reduction process described in (Rosenbrock 1970). Then the set

7
P P
difference of o.d. zeros of and the o.d. zeros of 1 is called the set of input output
P P
decoupling zeros (i.o.d. zeros) of . The set of i.d. zeros of defined by (6) contains the
P P
set of i.o.d. zeros of . Trivially, the set of o.d. zeros of defined by (7) contains the set
P
of i.o.d. zeros of .

At the same time, Rosenbrock (1970) introduced zeros of a transfer function matrix,
G(s). For a matrix G(s) put into Smith-McMillan form, the zeros of the numerator poly-
nomials, εi (s), are called the zeros of G(s) and the zeros of the denominator polynomials,
ψi (s), are called the poles of G(s). Because G(s) is called the transmittance matrix and
its zeros are physically associated with the transmission-blocking properties of the system
it describes, the zeros of G(s) are commonly called transmission zeros. Since the transfer
function matrix describes that part of a system which is both reachable and observable,
the sets of transmission zeros and decoupling zeros are conceptually distinct.

In addition to the previous definitions, Rosenbrock (1973, 1974b) later defined system
zeros. System zeros, loosely the combined set of decoupling and transmission zeros, are
P
defined using the Rosenbrock system matrix, . In a manner similar to his definition
P
of transmission zeros, Rosenbrock first defined the zeros of a system by putting into
Smith form. The zeros of a system were then defined to be the zeros of the polynomials
λi (s), 1 ≤ i ≤ r, taken all together (Rosenbrock 1973). However, this definition was later
revised in (Rosenbrock 1974b) in order to establish an exact set equality. The first definition
of system zeros, although not an appropriate definition to establish this set equality, was
in equivalent definitions designated as yielding invariant zeros, the roots of the invariant
polynomials, λi (s). Invariant zeros contain the complete set of transmission zeros and some,
but not necessarily all, of the decoupling zeros.
P
The revised definition of zeros of a system is based also upon the system matrix . If
P
has rank n + r where 0 ≤ r ≤ min(m, p), then consider the (n + r)-order nonzero minors
P
of of the form
P1,2,...,n,n+i1 ,n+i2 ,...,n+ir
1,2,...,n,n+j1 ,n+j2 ,...,n+jr .
P
Alternatively, consider that subset of minors of of maximum order containing the first
P
n rows and n columns of . Let φ(s) be the monic greatest common divisor of all of these

8
P
minors of . The system zeros are then the complex roots including multiplicities of φ(s).
As noted in (Rosenbrock 1974b), the set of invariant zeros and the set of system zeros
P P
coincide when m = p and det( ) 6= 0, that is, is nondegenerate. The set of invariant
zeros is a subset of the set of system zeros. The following set equality was shown to be true
by Rosenbrock:

{system zeros} = {transmission zeros}+{i.d. zeros}+{o.d. zeros}−{i.o.d. zeros}. (8)

The zeros of multivariable systems discussed up to this point have been considered to
be finite: zeros at finite points in the complex s plane. This is due to the fact that the
unimodular matrices U1 (s) and U2 (s) used to generate the Smith or Smith-McMillan forms
do not retain information regarding behavior at s = ∞ as they can modify poles and zeros
at infinity by both introduction and or deletion (Kailath 1980). In the process of defining
the degree of G(s), Rosenbrock (1970) extended his definitions of finite poles and zeros of
G(s). A pole at infinity corresponds to g(s) → ∞ as s → ∞ where g(s) is any element of
G(s). Similarly, a transmission zero at infinity corresponds to every k-order minor of G(s),
for some given k, approaching zero as s → ∞. A bilinear transformation
αp
s=
p−1
for α not equal to any finite pole or zero of any minor of G(s) allows G(s) to be written as
!
αp
G(s) = G = H(p).
p−1

Then G(s) has an infinite zero if and only if H(p) has a transmission zero at p = 1. For
additional theorems based on this idea, see Rosenbrock (1970).

The decoupling zeros previously defined by Rosenbrock (1970) were later termed finite
decoupling zeros to distinguish them from infinite decoupling zeros (see Rosenbrock 1974a).
The more general system under consideration is

E ẋ(t) = Ax(t) + Bu(t) (9)


y(t) = Cx(t)

9
with E singular, x(t)  Rn , u(t)  Rm , and y(t)  Rp . Note that (9) reduces to (1) for
E nonsingular. Infinite input decoupling zeros are equal in number to those finite input
decoupling zeros of
" #
E − sA B
−C 0

which are equal to zero. Similar results apply to both output decoupling and input output
decoupling zeros.

The study of zeros of linear multivariable systems has led to the solution of many control
problems. For example, properties of zeros have led to the design of stable high gain feed-
back systems (Shaked and Kouvaritakis 1977), distributed systems (Callier et al. 1981),
and nonminimum phase systems (Tsai and Wang 1987), and conditions for controllability
(Glover and Silverman 1976) or conditional reachability and observability (Suda and Mut-
suyoshi 1978). The physical output-zeroing problem (Karcanias and Kouvaritakis 1979,
Al-Nasr et al. 1981, Al-Nasr 1982) as well as characterization of fixed modes (Davison
and Özgüner 1983, Davison and Wang 1985, Tarokh 1985) have been approached using
zeros. Moreover, zeros have been used to resolve stability questions with respect to block
decoupling (Williams and Antsaklis 1986). The effect of the sampling rate in the sampling
of a continuous system on the resulting zeros of the sampled system has been examined
extensively (Åström et al. 1984, Passino and Antsaklis 1988, Fu and Dumont 1989). Also,
zeros of discrete systems are extremely important in adaptive control (Elliott and Wolovich
1982, M’Saad et al. 1985). A concentration on assignment or cancellation of zeros is also in
evidence in the literature (Murdoch 1973, Wolovich 1974, Seraji 1975, Patel 1978, Vardu-
lakis 1980a, Franklin and Johnson 1981, Cameron and Zolghadri-Jahromi 1982, Lee et al.
1987, Karcanias et al. 1988, Williams and Juang 1988, Misra and Patel 1988). Zeros are
used in model reduction (Shaked and Karcanias 1976) and in the selection of weights for LQ
regulator design (Harvey and Stein 1978). Additionally, zeros have been defined in systems
with delays (Pandolfi 1982), determined in spacecraft control (Ohkami and Likins 1976),
and investigated in large space structures (Williams 1989a,b). Zeros are intimately related
to the study of linear singular systems (Lewis 1986). Time-domain behavior due to zero
location has been extended to the multivariable case (Porter and Jones 1985). Theoreti-

10
cally, zeros are used in the hierarchical theory of systems (Rosenbrock and Pugh 1974) and
in greatest common divisor extraction of polynomial matrices (Silverman and Van Dooren
1979). Furthermore, zeros are central in the solution of model matching and factorization
problems (Wolovich et al. 1977, Özgüler and Eldem 1985, Antsaklis 1986, Sain et al. 1988)
and in the theory of inverse systems (Rosenbrock and Van Der Weiden 1977).

III. TRANSMISSION ZEROS

Within the topic of zeros of multivariable systems, the largest class of papers revolves
around the use of the transfer function matrix. Among these is a subset which defines
finite transmission zeros merely as specific complex numbers satisfying a given property.
Structural properties such as multiplicity or invariant factors are not evident within this
group.

A common approach involving zeros of transfer function matrices is the use of coprime
factorization. C.A. Desoer and J.D. Schulman (1974) presented a definition of transmission
zeros in terms of left coprime and right coprime factorization of a transfer function matrix.
Given any p × m transfer function matrix G(s), the following nonunique factorizations can
occur:

G(s) = Nr (s)Dr−1(s) = D`−1 (s)N` (s). (10)

The polynomial matrices Nr (s) and Dr (s) are relatively right prime and N` (s) and D` (s) are
relatively left prime with both Dr (s) and D` (s) nonsingular. Any p × m polynomial matrix
Nr (s) or N` (s) satisfying (10) is called a numerator of G(s). The zeros of G(s) are defined
via a drop in the local rank below normal rank of the numerator matrix, N` (·). These
authors have implied that G(s) must be of full rank. The transmission blocking properties
of these zeros are also interpreted dynamically, justifying the term transmission zero. The
order or multiplicity of such a zero cannot be determined using the above definition; instead,
it is noted that a canonical form of G(s) must be used, see also (Wolovich 1973b).

The definition of transmission zeros given in (Desoer and Schulman 1974) was general-
ized by A.G.J. MacFarlane (1975), who defined the zeros of G(s) to be that set of complex

11
numbers {s0 } which satisfy
rank G(s0 ) < rank G(·).

The definition in (MacFarlane 1975) follows immediately from the definition in (Desoer and
Schulman 1974), provided that no transmission zero coincides with a pole of G(s). However,
some difficulties with this definition in the determination of rank occur with coincident
zeros and poles creating infinite elements of G(s) (Pugh 1977). Again, multiplicity of zeros
cannot be determined.

Such structural properties as order of a transmission zero prompted other investigations.


In much the same way as Rosenbrock’s definition of transmission zeros (zeros of a transfer
function matrix), many authors choose not only to use the transfer function matrix as a
means of investigating multivariable zeros, but also to utilize the Smith-McMillan form of
the transfer function matrix.

B.A. Francis and W.M. Wonham (1975) in their paper involving the role of transmis-
sion zeros in regard to linear multivariable regulators use the transmission zero as defined
by Rosenbrock. Previously, Davison and Wang (1974) had indicated the significance of
the ”transmission” zero for the solvability of a general regulator problem. Francis and
Wonham chose the Rosenbrock definition because of its accommodation of multiple zeros,
a crucial point in their study. Compensators which regulate systems employ the multi-
variable analogue of pole-zero cancellation and, to function in the presence of variations in
system parameters, must supply transmission zeros in greater multiplicity.

¿From a strictly state space point of view, A.S. Morse (1973) identified some structural
invariant properties of the matrix triple (A, B, C). The transmission polynomials, the roots
of which are the transmission zeros of (A, B, C) are described. It is believed that the term
”transmission zero” was first defined in a system theoretic context here. The author was
inspired by E.A. Guillemin (1957) who used the term in reference to those values of s at
which the value of a transfer impedance is zero (Morse 1989). A proof that this set of
transmission polynomials is equivalent to the set of numerator polynomials, {εi (s)}, of the
Smith-McMillan form of the system’s transfer function matrix was presented by B.C. Moore
and L.M. Silverman (1974). Alternate proofs were given by S. Hosoe (1975), Corfmat and

12
Morse (1976), and B.D.O. Anderson (1976).

The zeros of the Smith-McMillan form of a transfer function matrix were also found
to be equivalent to those zeros of a certain class of multivariable systems defined by T.
Mita (1977). The concept of Maximal Unobservable Subspace (MUS) is introduced and
only the poles of the MUS which are invariant under state feedback are shown to be the
transmission zeros defined by Rosenbrock. Furthermore, the stability conditions of zero-
output systems and zero-sensitive systems are investigated using this definition and give
new interpretations to the zeros of the Smith-McMillan form.

In response to Rosenbrock, W.A. Wolovich (1973b) presented the definition of a ”nu-


merator” of a rational transfer function matrix as a natural extension of the SISO case.
Any two numerators of a p × m rational transfer function matrix G(s) as in (10) are shown
to be equivalent; i.e.
UL (s)Nr (s)UR (s) = N` (s)

for an appropriate pair of unimodular matrices, UL (s) and UR (s). Several equivalent defi-
nitions of zeros of transfer function matrices were presented including that of Rosenbrock.
Particularly, if a numerator of G(s) has rank r, then the zeros of G(s) are equal to the
zeros of each of the (r) nonzero diagonal elements λi (s) of the Smith form of the numerator
matrix. It was noted that the zeros of G(s) are those zeros common to all of the nonzero
r-order minors of any numerator of G(s) of rank r. If any numerator of G(s) is reduced to
triangular form, then only one r-order minor will be nonsingular, and therefore the zeros
of G(s) will be given by the zeros of the main diagonal elements of the numerator matrix.

In a manner similar to (Wolovich 1973b), Desoer and Schulman (1974), in addition to


their aforementioned rank drop definition, indicated that the Smith-McMillan form of G(s)
can be used to include multiplicity information. In addition, the transmission zeros were
shown to be the poles, including multiplicity, of a reflexive generalized inverse (Boullion
and Odell 1971), a natural extension of the SISO case.

In 1975, R.V. Patel described the zeros of a transfer function matrix of an invertible

13
system (F, G, H)

ẋ(t) = F x(t) + Gu(t) (11)


y(t) = Hx(t)

with x(t)  Rn , u(t)  Rm , and y(t)  Rp , and showed that these transmission zeros were
equivalent to the transmission zeros of a [min(n − m, n − p)]-order, m-input, p-output
system (A, B, C, D). The transfer function matrix corresponding to (11) is

W (s) = H(sI − F )−1 G,

which is factored into left or right coprime factorizations as in (Desoer and Schulman 1974,
Wolovich 1973b) to produce the lower-order system. Furthermore, the invertible system
(F, G, H),
rank W (s) = min(m, p),

is either left invertible (p > m), right invertible (p < m), or normally invertible (p = m).
Minimal order right or left inverses (WR (s) or WL (s)) are used (those inverses with the
lowest order characteristic polynomial) as in (Desoer and Schulman 1974, Wolovich 1973b)
to determine that the zeros of W (s) are the poles of WR (s) or WL (s), the minimal order
inverses.

Another method of defining transmission zeros with reference to the Smith-McMillan


form of G(s) is through the use of the minors of G(s), the dual of which is used in a
definition of poles; this was noted in (MacFarlane and Karcanias 1976). The zeros of a
G(s) of rank r are the roots of the greatest common divisor of the numerators of all the
r-order minors, provided that all these minors were adjusted to have the pole polynomial
of G(s),
r
Y
ψi (s),
i=1

as their common denominator. Additionally, the McMillan degree of G(s) was found to
determine an upper bound on the number of transmission zeros (Pugh 1977).

Several approaches to transmission zero definition exist other than via the Smith-
McMillan form or coprime factorization. As a response to Desoer and Schulman (1974),

14
J. Vandewalle and P. Dewilde (1974), provided a simple formula for the order and degree
of a multivariable zero at a point p0 using the Laurent expansion of the transfer function
matrix at p0 , thereby avoiding the need for computing the inverse of a matrix. Comparison
to results using the Smith-McMillan form of G(s) supports the proof.

A method for defining zeros of an m × m transfer function matrix G(s) resulted from
a study by A.G.J. MacFarlane and I. Postlethwaite (1977) on the generalized Nyquist
stability criterion for multivariable systems. The poles and zeros of G(s) are defined using
characteristic functions and define subsets of both poles and transmission zeros defined
algebraically in (Rosenbrock 1970). Motivated by this algebraic function approach, M.C.
Smith (1981) used coprime factorization of the transfer function matrix in order to develop a
relationship between the number of algebraic function zeros and the number of transmission
zeros found using the Smith-McMillan form. Further investigations with respect to zeros
defined via algebraic functions can be found in (Smith 1984, 1986).

MacFarlane was also instrumental in the development of the geometric approach to


multivariable zeros. B. Kouvaritakis and A.G.J. MacFarlane (1976a,b) presented their
ideas in a two-part series of papers. These papers geometrically interpreted locations
of invariant zeros for both square and nonsquare transfer function matrices. All state
space system models involved were assumed to be completely controllable and observable;
therefore, the terms invariant zero, transmission zero, and zero were used interchangeably.

A direct consequence of these ideas can be found in the determination of zeros and their
corresponding zero directions. Associated with an invariant zero si , there can exist nonzero
P
vectors xi  Rn and ui  Rm such that the augmented vector [xi ui ]0 lies in the kernel of
evaluated at si , i.e.
" # " #
si I − A B xi
= 0n+p .
−C D −ui

The vector xi (ui) is called a state (input) direction and the vector [xi ui ]0 is called an
invariant zero direction or zero direction corresponding to si . Physically, applying an input
of the form
u(t) = ui exp(si t)1(t),

15
where
( )
1 t≥0
1(t) =
0 t<0

will, if si is not a pole of the system, produce a state of the form

x(t) = xi exp(si t), t ≥ 0

and output of the form


y(t) = 0, t ≥ 0.

This problem has been addressed as the output-zeroing problem in the literature (MacFar-
lane and Karcanias 1976).

V. Lovass-Nagy and D.L. Powers (1980) proposed a method to determine zeros and
zero directions using matrix generalized inverses. Expressions for the direct calculation of
transmission zeros were included.

Transmission zeros at infinity have received considerable attention recently. Infinite


transmission zeros are important in determining the asymptotic behavior of multivariable
root loci (MacFarlane et al. 1978, Kouvaritakis and Edmunds 1979, Hung and MacFar-
lane 1981, Smith 1986). They have also been investigated in the light of the problems of
decoupling (Vardulakis 1980b), realization (Bosgra and Van Der Weiden 1981), and model
matching (Malabre and Kučera 1984).

In an approach similar to that of Rosenbrock (1970), several authors have defined infinite
transmission zeros using the Smith-McMillan form of G(s). A transfer function matrix
G(s) is said to have an infinite zero of degree k if w = 0 is a finite transmission zero
of degree k for the rational matrix G(1/w); this infinite transmission zero definition is
evaluated and compared to that of Rosenbrock (1970) by A.C. Pugh and P.A. Ratcliffe
(1979). For strongly irreducible systems of the form (9), the infinite zero structure of G(s)
P
was found to be isomorphic to the infinite zero structure of (Verghese et al. 1979).
Infinite transmission zeros were shown to be invariant under constant output and state
feedback by P.M.G. Ferreira (1982), although this invariance can be extended to the more
general case of dynamic compensation via the interactor (Wolovich and Falb 1976, Antsaklis

16
1986). Because unimodular matrices used in creating the Smith-McMillan form of G(s) can
alter zero information at infinity, A.I.G. Vardulakis, D.N.J. Limebeer, and N. Karcanias
(1982) proposed a set of bi-proper elementary operations instead of unimodular operations
to compute the Smith-McMillan form of a transfer function matrix at infinity, thus retaining
information necessary for the calculation of infinite transmission zeros.

The geometric approach has been used to study transmission zeros at infinity. C.
Commault and J.M. Dion (1982) described the infinite zero structure geometrically using
the concepts of almost (A, B)-invariant and almost controllability subspaces. This quite
formal geometric approach is very different in spirit from that used by Wonham (1974)
in relation to finite zeros. A connection with the infinite transmission zero in (Pugh and
Ratcliffe 1979) was noted. Both Kouvaritakis and MacFarlane (1976a,b) and Kouvaritakis
and Shaked (1976) established a geometric method to find infinite transmission zeros which
Shaked (1977) used to establish that all root loci asymptotes of passive, linear, time-
invariant systems at infinity are in the direction of the negative real axis.

IV. INVARIANT ZEROS

The concentration on the state space description matrices (A, B, C, D) as well as the
P
system matrix in defining multivariable zeros led directly to the invariant zero. Nearly
coincident with Rosenbrock’s definition of invariant zero (1973), W.A. Wolovich (1973a)
proposed that zeros of a linear, time-invariant, multivariable system be defined via a system
matrix rank test. The state space representation under consideration was

ẋ(t) = Ax(t) + Bu(t)


y(t) = Cx(t) + E(D)u(t)

with D = d/dt, E(D) a polynomial matrix in D, and x(t)  Rn , u(t)  Rm , y(t)  Rp . The
zeros of a minimal system were defined as the complex numbers {s0 } which satisfy
" # " #
−s0 I + A B −sI + A B
rank < rank .
C E(s0 ) C E(s)

17
Furthermore, these zeros were said to exist if and only if they were also zeros of the cor-
responding transfer function matrix G(s). This statement was qualified by the distinction
between minimal and nonminimal zeros (identifying those modes corresponding to input or
output decoupling zeros) for a more general, possibly nonminimal, system. The invariant
zeros are both the minimal and nonminimal zeros, the complex s0 which reduce the normal
P
rank of . The minimal zeros are the previously defined transmission zeros.

Shortly thereafter, E.J. Davison and S.-H. Wang (1974) introduced a definition of
”transmission” zeros because of their importance in regulation problems, decoupling the-
ory, and servomechanisms. The ”transmission” zeros of (1) are the complex numbers {s0 }
which satisfy
" #
A − s0 I B
rank < n + min(m, p).
C D

This definition assumes that (A, B, C, D) is nondegenerate. A system (A, B, C, D) is defined


to be degenerate if, for any specified n+1 distinct scalars λj  R, 1 ≤ j ≤ n+1, the following
relationship is true:
" #
A − λj I B
rank < n + min(m, p).
C D

If (1) is controllable and observable, then the ”transmission” zeros here defined are equiv-
alent to Rosenbrock’s transmission zeros. If (1) is nondegenerate, then the ”transmission”
zeros here defined are Rosenbrock’s invariant zeros. This definition is similar to that of
Wolovich (1973a). The paper also discussed ”transmission” zeros of cascade systems and
how they relate to the poles of the closed loop system under high gain output feedback.
Nonsquare systems were also shown to have ”transmission” zeros only rarely.

The degeneracy issue, zeros everywhere in the complex plane, was exactly characterized
by G.A. Hewer and J.M. Martin (1984) in terms of system matrices. The idea of non-
degeneracy was linked to the concept of functional reproducibility. Davison and Wang’s
definition yields an infinite set of ”transmission” zeros for a degenerate system. The treat-
ment by Wolovich produces a finite set for a degenerate system. In both of these cases,
the definitions are inadequate in dealing with the case of repeated zeros and will compute

18
not only transmission zeros, but also some decoupling zeros. The term ”transmission” zero
may therefore not be appropriate; invariant zero is more appropriate. See also the notion
of extended or generic zeros in Section VI.

Some definitions of invariant zeros have addressed structural questions. In addition to


presenting the relationship between algebraic and geometric zeros, MacFarlane and Karca-
nias (1976) defined invariant zeros to be the zeros of the monic greatest common divisor of
P
all maximum-order minors of . In response, H.H. Rosenbrock (1977) in a correspondence
item proved that these newly defined invariant zeros were in fact the set of zeros specified
in his first definition of system zeros (Rosenbrock 1973).

R.V. Patel (1976) extended Ilia Kaufman’s method of zero computation for SISO sys-
tems (1973) in order to provide a method for the computation of invariant zeros of invertible
multivariable systems with square transfer function matrices (p = m). Because the systems
considered are invertible, the m × m transfer function matrix G(s) has rank m and the
following calculations can be made:

detG(s) = det[C(sI − A)−1 B + D]


P
det
=
det(sI − A)
p(s)
= . (12)
q(s)
The invariant zeros are said to be the zeros of p(s) in (12) because p(s) = 0 precisely
P
when loses rank. Again, coincident poles and zeros can create an ambiguity. Allowing
no cancellations between coincident poles and zeros in (12) produces invariant zeros. If
the system is minimal and if no cancellations between p(s) and q(s) are allowed, then
(12) computes transmission zeros. If p(s) and q(s) are reduced so that the numerator and
denominator of (12) are coprime, then a subset of transmission zeros is calculated. Clearly,
detG(s) does not contain the same zero information as the Smith-McMillan form of G(s).

B.P. Molinari (1976) described the invariant zeros in terms of reachability and observ-
ability. Based upon the invariant zero definition of Rosenbrock (1973), this paper merely
connects system properties and (A, B)-invariant subspaces with these zeros. P. Sannuti and
A. Saberi (1987) developed a special coordinate basis to address some of the fundamental

19
properties of linear systems, one of these being invariant zeros.

W.M. Wonham’s geometric approach to linear multivariable control (1974) related in-
variant zeros to the supremal controllability subspace contained in the kernel of C, R∗ , and
the supremal (A, B)-invariant subspace contained in the kernel of C, V ∗ . Note that MUS
defined in (Mita 1977) is, in fact, V ∗ . Using minimal system inverses, G. Bengtsson (1974)
adopted these geometric concepts and established invariant transmission zeros as a subset
of the poles of a left (right) inverse system, if one exists. Under certain conditions, J.P.
Corfmat and A.S. Morse (1976) identified transmission zeros in terms of these subspaces.
The number of invariant zeros has been shown to aid in the determination of the dimensions
of V ∗ and R∗ (Antsaklis 1980, Antsaklis and Williams 1980). These papers extend similar
results of Mita (1977) as well as discuss the role of decoupling zeros in this problem. For
a class of discrete spectral systems, the invariant zeros of the transfer function have been
shown to characterize invariant subspaces (Zwart 1988).

The concept of invariant zero directions, initially developed in (MacFarlane and Karca-
nias 1976), was established in conjunction with MacFarlane and Karcanias’s invariant zero
to characterize the output-zeroing problem. U. Shaked and N. Karcanias (1976) utilized
this theory in model reduction, while B. Kouvaritakis (1976) examined the duality between
poles and pole directions and zeros and zero directions using inverse systems. D.H. Owens
(1977) established the geometric multiplicity of an invariant zero as the dimension of the
subspace generated by its corresponding state zero directions. Considering slow and fast
subsystems describing slow and fast modes of a system, B. Porter (1978) determined slow
and fast invariant zeros and their corresponding zero directions. Lovass-Nagy and Powers
(1980) determined the invariant zeros and zero directions using matrix generalized inverses.

In addition to finite invariant zero investigations, the topic of infinite invariant zeros
has been addressed by Bosgra and Van Der Weiden (1981). The invariant zeros at infinity
were defined, including multiplicity.

20
V. SYSTEM ZEROS

Recall that invariant zeros contain all of the transmission zeros, but not necessarily
all of the decoupling zeros. However the system zeros, as defined by Rosenbrock (1974b),
contain all transmission and decoupling zeros. The system zeros were defined to satisfy
the set equality in (8), while the invariant zeros have been found to contain all of the
transmission zeros, all of the o.d. zeros, and some of the i.d. zeros which are not i.o.d. zeros
(Rosenbrock 1977).

Sannuti and Saberi (1987) defined system zeros in a similar manner using their special
coordinate basis. Again, system zeros were defined in set equality form using invariant
and decoupling zeros. In (1973, 1974b), Rosenbrock addressed the duality of poles and
zeros of a system using poles and zeros of G(s); the set of system zeros (poles) is obtained
from the union of the set of zeros (poles) of G(s) and the same set of decoupling zeros, see
also (MacFarlane and Karcanias 1976, Pugh 1977). Another set characterization of system
zeros was found in (Owens 1977). The set of system zeros of (A, B, C) was described as
the intersection over all m × p constant matrices K of the invariant zeros of (A, B, KC).
P
System zeros as well as invariant zeros have been described using minors of in order
to address structural questions. Rosenbrock’s definition of system zero (1974b) was used by
Pugh (1977) to establish an upper bound on the number of system zeros in a way analogous
to classical control theory. The system zeros of (A, B, C), including multiplicities, were
also defined by Owens in 1977 to be the zeros of the monic greatest common divisor of all
polynomials of the form
" #
sI − A −B
det
KC 0
for any m × p constant matrix K.

System zeros by their very nature are not used in conjunction with invariant zero direc-
tions; however, system zeros at infinity have been approached by several authors. Bosgra
and Van Der Weiden (1981) defined system zeros at infinity using minors. An interest-
ing extension of Rosenbrock’s system zero definition was provided in P.M.G. Ferreira’s
definition of infinite system zero (1980). It was concluded that the

21
number of infinite system zeros = number of infinite transmission zeros
+ number of infinite i.d. zeros
+ number of infinite o.d. zeros
− number of infinite i.o.d. zeros.

Infinite system zeros were found to be invariant under constant output feedback, and also
P
invariant for constant state feedback for of normal rank n + m (Ferreira 1982).

VI. ZEROS AND MODULES

We set the following conventions. For a field k, let k[s] be the ring of polynomials in
s with coefficients in k, and let k(s) be the induced quotient field. Recall that a ”transfer
function” in k(s) is an equivalence class, and note that the customary representative of a
class is a pair (n(s), d(s)) of polynomials in k[s], with d(s) nonzero and the pair relatively
prime.

For a discussion of poles and zeros in the multivariable sense, it is necessary to have
in hand a concrete representation of polynomials and transfer functions in a vector sense.
Intuitively, we do this by ”multiplying” the scalar version of the quantity in question onto
the vector of interest. More precisely, let U and Y be finite dimensional vector spaces over
k; and notice that k[s] and k(s) are also k-vector spaces. Then we employ the k-bilinear
tensor product to form k[s]-modules

U[s] = k[s] ⊗k U, Y [s] = k[s] ⊗k Y,

and k(s)-vector spaces

U(s) = k(s) ⊗k U, Y (s) = k(s) ⊗k Y.

If G(s) : U(s) → Y (s) is a k(s)-linear map, we wish to describe its poles and zeros in
a precise, technical sense. For the elementary class represented by (n(s), d(s)), this is
done easily and intuitively with the k[s]-modules k[s]/d(s)k[s] and k[s]/n(s)k[s]. For G(s),
greater elaboration is needed.

22
The pole module X(G(s)) associated with a k(s)-linear map G(s) is defined to be the
k[s]-factor module given by

X(G(s)) = U[s]/{U[s] ∩ G−1 Y [s]}, (13)

where G−1 is the inverse image function of G(s), defined on Y [s] by

G−1 Y [s] = {u(s) : u(s)  U(s) and G(s)u(s)  Y [s]}.

As a k[s]-module, X(G(s)) is finitely generated, because U[s] is finitely generated; it is a


torsion module because every element in U[s], even if not in G−1 Y [s], can be made so by
scaling it with a suitable polynomial p(s) in k[s].

Finitely generated, torsion modules over k[s], such as the pole module X(G(s)), can be
understood as traditional state spaces of finite dimension over the field k. Considered as
a k-vector space, X(G(s)) = XP , the usual state space of a minimal realization of G(s);
and the scalar module action s : X(G(s)) → X(G(s)) defines a k-linear dynamics map
AP : XP → XP in the realization.

The finitely generated, torsion zero module Z(G(s)) associated with a k(s)-linear map
G(s) was defined by Wyman and Sain (1981a) to be the k[s]-factor module

Z(G(s)) = {G−1 Y [s] + U[s]}/{kerG(s) + U[s]}. (14)

In a manner analogous to poles, the zero module permits a state space interpretation.
As a k-vector space, Z(G(s)) = XZ , which is of finite dimension. Also, the scalar module
action s : Z(G(s)) → Z(G(s)) defines a k-linear map AZ : XZ → XZ . The space XZ
is not the usual state space; and the map AZ is not the usual dynamics map. In order to
distinguish the cases, we can think of XP as a space of pole-states and of XZ as a space
of zero-states. Then AP becomes the pole-state dynamics map, while AZ is the zero-state
dynamics map.

The relationship between the zero module (14) and invariant subspace approaches of
Wonham (1974) and others was initially studied in (Wyman and Sain 1980). Zero modules
can be employed to study the pole structure of inverse systems (Wyman and Sain 1981b,

23
1985). Moreover, the same family of methods has been employed to examine pole and zero
character in solutions to factorization problems (Conte, Perdon, and Wyman 1988, Sain et
al. 1988). As technical tools, the pole and zero modules have been useful in exploring the
very complicated question of multivariable pole-zero cancellation (Wyman and Sain 1981c,
Conte and Perdon 1985a, 1986, Conte et al. 1986). Zero modules have also been formed for
system matrices (Horan 1980, Wyman and Sain 1987). By extension to rings other than
k[s], the zero module has been employed to study zeros at infinity (Conte and Perdon 1984,
1985b). The geometric approach of Conte and Perdon (1984) for zeros at infinity, zeros
defined by means of invariant factors of modules which come from invariant subspaces, is
similar to that of Wonham (1974) for finite zeros.

When multiple inputs and multiple outputs are considered, however, zeros may appear
which are not of pole-type. Some of these are torsion zeros, but not finitely generated; in
the context of realization theory, these may be classified as zeros of output type. Others
are finitely generated, but free; and these have an analogous classification as zeros of input
type. It is possible to extend the idea of zeros in such a way as to embrace both the zeros
of pole-type, and zeros of output or input type.

We begin our discussion of extended zeros in such a way as to address the zeros of
output type. Consider the k[s]-factor module

ZΓ (G(s)) = G−1 Y [s]/{U[s] ∩ G−1 Y [s]}. (15)

This module is torsion. Indeed, observe that G−1 Y [s] is a submodule of U(s), from which
we have for every representative r(s) of a class in (15) a polynomial p(s) in k[s] such that
p(s)r(s) is in U[s] and is therefore equivalent to zero. If kerG(s) is not equal to zero,
however, (15) is not finitely generated. We shall call (15) the Γ-zero module of G(s). Now
let
Γ(G(s)) = kerG(s)/{U[s] ∩ kerG(s)}. (16)

The module (16) is both torsion and divisible. To see the latter, observe that scalar
multiplication p(s) : kerG(s) → kerG(s) is an epimorphism of k[s]-modules whenever
p(s) is nonzero in k[s]. Then Γ(G(s)) inherits this same property. As a divisible module

24
over k[s], Γ(G(s)) is injective; and so its image under injection into ZΓ (G(s)) is a direct
summand. It follows that

ZΓ (G(s)) ≈ Γ(G(s)) ⊕ Z(G(s)). (17)

We will refer to Γ(G(s)) as the divisible zero module of G(s), and to ZΓ (G(s)), the Γ-zero
module of G(s), as an extended zero module. In view of (17), we see that the notion of
Γ-extended zeros includes both Γ(G(s)) and Z(G(s)) as special cases.

Next we examine zeros of input type, by forming the k[s]-factor module

ZΩ (G(s)) = Y [s]/{Y [s] ∩ GU[s]}. (18)

Module (18) is finitely generated, because Y [s] is finitely generated. Not every equivalence
class in (18) is torsion, however, because not every y(s) in Y [s] can be brought into GU[s]
by means of scalar multiplication. If the image, imG(s), of G(s) is equal to Y (s) on the
k(s)-level, then Z(G(s)) is equal to (18); and so we see that ZΩ (G(s)) differs from Z(G(s))
when cokerG(s) is nonzero. We call (18) the Ω-zero module of G(s). If

Ω(G(s)) = Y [s]/{Y [s] ∩ GU(s)}, (19)

then it can be shown that

ZΩ (G(s)) ≈ Ω(G(s)) ⊕ Z(G(s)). (20)

We will call Ω(G(s)) the free zero module of G(s). Like ZΓ (G(s)), then, ZΩ (G(s)) is an
extended zero module, this time of Ω-type. Notice from (20) that extended Ω-zeros include
Ω(G(s)) and Z(G(s)) as particular cases.

Extended zero modules have been shown to play a central role in relating to invariant
subspace theory and transmission blocking (Wyman and Sain 1983), as well as model
matching (Sain et al. 1988). For system matrices, extended zero modules have been defined
and related to input and output decoupling zero modules (Wyman and Sain 1982, 1987).
In addition, extended zero modules are shown to be important in a complete accounting of
the poles and zeros of a transfer matrix (Wyman et al. 1988, 1989).

25
The divisible or free zeros, when considered on the system level, offer a concrete and
specific technical characterization of the degeneracy issues described in Section IV. For this
reason, we can also refer to them as modules of generic zeros.

VII. FURTHER ZERO DIRECTIONS

Because of space limitations, we can only mention here a number of additional ideas
and trends. Other zeros besides those previously mentioned have been defined and shown
to be useful in the solution of control problems.

The term blocking zero was first introduced by P.G. Ferreira and S.P. Bhattacharyya
(1977) as a result of the systematizing of the design of tracking systems. In general,
these blocking zeros differ in multiplicity and value from the zeros defined by Rosenbrock
and are, in fact, a subset of those zeros defined in (Rosenbrock 1970, 1974b). This same
term, blocking zero, was later used by R.V. Patel (1986) and was shown to be related to
the uncontrollability and or unobservability of (A, B, C) and therefore also was related to
decoupling zeros. A correspondence item and rebuttal regarding this ambiguous use of
terminology can be found in (Ferreira 1986). It should be noted that these two definitions
of blocking zeros are not equivalent.

Another new definition of zeros of a multivariable system, disturbance zeros, appeared


in (Patel et al. 1977). Instead of relating outputs to inputs, these zeros relate outputs to
disturbances. Disturbance zeros are found to be invariant under output feedback but not
state feedback and can therefore be assigned using state feedback.

Emami-Naeini and Rock (1984) defined compromise zeros in an investigation of the


asymptotic behavior of the closed loop poles of nonsquare multivariable systems. The
compromise zeros refer to the locations, approached by closed loop system poles under
high gain feedback, which are not the locations of minimum phase transmission zeros, the
reflections of nonminimum phase transmission zeros, or infinity. For example, if a system
has more outputs than controls,the presence of compromise zeros illustrates a restriction
on the system’s ability to track every output.

26
The definition of a model matching transformation zero as a rank deficiency condition
which prevents the transformation of an H 2 or H ∞ optimal control problem into a corre-
sponding model matching problem is introduced in (O’Young et al. 1989). Methods for the
elimination of such zeros are also discussed. Similarly, in regard to H ∞ control design, the
computation of the H ∞ norm for systems with multiple right half plane zeros is treated in
(Chang and Banda 1989).

Invariant and transmission zeros of subsystems, respectively termed invariant and trans-
mission subzeros, have been defined recently (Schrader and Sain 1989). Theoretical subzero
movement is characterized with respect to feedback gain. Thus, plants may be precondi-
tioned to meet design specifications on transient response.

The module zero approach has produced the definition of a dual zero space (Wyman
and Sain 1986). The study of state spaces for left and right inverse systems is therefore
unified, as is the treatment of pole and zero questions in inverse systems.

The importance of the infinite zero structure with respect to the problem of decoupling is
noted in (Dion and Commault 1988, Amin and Moness 1988). Moreover, J.W. Grizzle and
M.H. Shor (1988) analyze the structure at infinity of a sampled-data system to determine
the effects of time-sampling on disturbance and input-output decoupling.

Recently, graph-theoretic tools have been used in zero investigations. B. Maione and B.
Turchiano (1988) provide necessary and sufficient conditions for the existence of decoupling
structural zeros with a digraph theory approach.

Another interesting area which relies on zero information is that of interpolation. Basic
interpolation problems for rational matrix functions are examined using zero and pole infor-
mation (Ball, Gohberg, and Rodman 1989). Furthermore, rectangular matrix polynomials
can be constructed from their zero data (Ball, Cohen, Rodman 1989).

Linear time varying systems have been examined from a zero viewpoint. Zeros of SISO
linear time varying systems have been defined with emphasis on the discrete-time case
(Kamen 1988a,b). Recently, a special class of such systems, discrete-time periodic systems,
has been examined (Conte, Longhi, and Perdon 1988) in the multivariable case in regard

27
to the structure at infinity, and, thus, the zeros at infinity. A study of module results with
respect to zeros for SISO discrete systems over a fixed finite field, their related periodic
behavior, and their connection to algebraic coding theory has been produced (Smithberger
and Wyman 1988). The concept of the zero module for multivariable systems over rings
has also been investigated (Conte and Perdon 1983).

Nonlinear as well as linear systems have been investigated yielding notions of nonlinear
multivariable zeros. Both finite and infinite nonlinear zero definitions have been approached
using the ideas of zero-dynamics and the zero-dynamics or structure algorithms (Byrnes
and Isidori 1984, 1985, 1988a,b, Isidori and Moog 1988, Di Benedetto and Slotine 1988,
Moog 1988). Alternative investigations regarding finite nonlinear zeros (Krener and Isidori
1980, D’Andrea and Praly 1988) and infinite nonlinear zeros (Nijmeijer and Schumacher
1985) have also occurred in the literature. In addition to their importance in decoupling
and stabilization problems, nonlinear zeros have been examined recently with regard to
global tracking (Barnard 1988) and flexible space platforms (Kwatny and Bennett 1988).

VIII. COMPUTATION

Numerical computation of zeros of multivariable systems has been a topic of consid-


erable interest. Although some aforementioned publications have provided methods of
computation of multivariable zeros, the following papers are predominantly concerned with
algorithms for calculation. Emphasis began with the calculation of zeros of individual el-
ements (transfer functions) of the transfer function matrix (Davison 1970). In response,
K.E. Bollinger and J.C. Mathur (1971) presented a numerical method to determine the
zeros of a transfer function. S.A. Marshall (1972) quickly followed suit with his related
transfer function technique. Other computational methods for transfer function zeros were
introduced in (Guidorzi 1972, Kaufman 1973, Guidorzi and Terragni 1973, Emami-Naeini
and Van Dooren 1982b). In response to Bollinger and Mathur (1971), Davison (1970), and
Guidorzi (1972), Patel (1973) proposed another computational method for calculation of
the numerator of a transfer function relating a single output to a single input.

28
Computational methods were expanded to include multivariable zeros. The concept of
zeros identified as poles of inverse systems prompted algorithms for transmission and invari-
ant zero calculation (Sinswat et al. 1976, Sinswat and Fallside 1976, 1977). However, the
transmission zeros computed therein are only a subset of Rosenbrock’s transmission zeros.
Methods based on the invariance property of zeros under high gain feedback (Kalnitsky
and Kwatny 1977, Young et al. 1977, Pohjolainen 1981) and the solution of a generalized
eigenvalue problem (Patel 1976) are available. Computation of (A, B)-invariant and con-
trollability subspaces is addressed in (Van Dooren et al. 1979, Moore and Laub 1978, 1979)
and calculation of zeros through a geometric approach in (Kouvaritakis and MacFarlane
1976a,b, Lindner 1979, Araki 1983). Decoupling and invariant zero calculation in addition
to invariant subspace computation using a unified matrix pencil approach has been exam-
ined recently (Demmel and Kågström 1987, 1988). Zero directions and their corresponding
zeros can be calculated (El-Ghezawi et al. 1982, Kouvaritakis and Daniel 1984, Amin and
Hassan 1988).

Frequently cited papers on numerical computation of zeros of multivariable systems were


by E.J. Davison and S.-H. Wang (1974, 1976, 1978), A.J. Laub and B.C. Moore (1978),
and A. Emami-Naeini and P. Van Dooren (1982a). Further comments on the Davison and
Wang techniques can be found in (Axelby and Davison 1976, Axelby et al. 1978). In the
P
treatments of Davison, Wang, Laub, and Moore, the Smith zeros of the system matrix
are called the ”transmission” zeros of the system. Recall that these authors are restricted
to the nondegenerate case. A widely known technique of Davison and Wang (1978) utilizes
the invariance property of zeros under high gain feedback. Given a nondegenerate system
(A, B, C, D) as in (1), then for ”almost all” full rank K and large ρ, the ”transmission”
zeros of (A, B, C, D) are a subset of the finite eigenvalues of

M(ρ) = A + BK((I/ρ) − DK)−1 C. (21)

Davison and Wang show that a nondegenerate (A, B, C, D) produces a nondegenerate


(A, BK, C, DK) for almost all K, and that the zeros of (A, B, C, D) are contained within
the zeros of (A, BK, C, DK). As the value of ρ approaches infinity, and if the matrix
((I/ρ) − DK) is invertible, the eigenvalues of M(ρ) in (21) converge to the zeros of

29
(A, BK, C, DK) or to infinity. The ”transmission” zeros are the unchanging eigenvalues
of M(ρ) when (21) is calculated with different values of ρ or different K. Those changing
eigenvalues are extraneous values.

The algorithm presented by Laub and Moore (1978) is restricted also to nondegenerate
systems. The square and nonsquare system matrix cases are approached in a slightly
different manner. For the square case, the numerically stable QZ algorithm (Moler and
Stewart 1973) is used to calculate the generalized eigenvalues of the square invertible system
P P
matrix which are the zeros of the system because is a regular matrix pencil. For
P
the nonsquare case, random rows or columns are used to pad the system matrix to
obtain a square pencil. This procedure is implemented two separate times and the common
eigenvalues between trials are the zeros of the system.

Finally, the technique of Emami-Naeini and Van Dooren (1982a) is based upon the
Kronecker canonical form of the system matrix. After a preliminary reduction into specified
form, the QZ algorithm is applied to a portion of the matrix which is a regular pencil whose
P
eigenvalues are the finite zeros of .

IX. CONCLUSIONS

When studying zeros of multivariable systems, one must specify which definition of
multivariable zero pertains to the problem at hand. Individual element zeros in a transfer
function matrix are related to transmission zeros only in special cases such as G(s) diagonal.
When a system is controllable and observable, the set of system zeros, the set of invariant
zeros, and the set of transmission zeros all coincide. Generally, these sets satisfy the
following relationships:

{system zeros} = {transmission zeros} + {i.d. zeros} + {o.d. zeros} − {i.o.d. zeros};

{system zeros} ⊃ {invariant zeros} ⊃ {transmission zeros}.

30
Investigations into multivariable zeros consider both the system and transfer function
matrix as a means of obtaining zero information. Zeros have also been approached through
the use of modules. Those definitions which consider zeros merely as complex numbers,
although deceivingly simple, create difficulties and produce little structural information.
The use of invariant factors includes multiplicity information while zero directions intro-
duce further insight into the output-zeroing problem. The module theory setting virtually
provides all zero information in one concise package.

ACKNOWLEDGEMENT

The authors wish to acknowledge a number of useful contributions in the transformation


of this document from the original draft. We thank our colleagues who have been so kind
as to read our manuscript and respond with helpful, deep, and historical suggestions and
comments.

REFERENCES

Al-Nasr, Nazar, ”On Zeros of Time-Invariant Multivariable Linear Systems Containing


Input Derivatives,” Int. J. Contr., Vol. 35, No. 4, pp. 749-753, April 1982.
Al-Nasr, Nazar, Lovass-Nagy, Victor, and Powers, David L., ”On Transmission Ze-
ros and Zero Directions of Multivariable Time-Invariant Linear Systems with Input-
Derivative Control,” Int. J. Contr., Vol. 33, No. 5, pp. 859-870, May 1981.
Amin, M.H., and Hassan, M.M., ”Determination of Invariant Zeros and Zero Directions
of the System S(A,B,C,E),” Int. J. Contr., Vol. 47, No. 4, pp. 1011-1041, April 1988.
Amin, M.H., and Moness, M., ”Design of Precompensators via Infinite-zero Structure
for Dynamic Decoupling of Linear Invertible Systems,” Int. J. Contr., Vol. 47, No. 4,
pp. 993-1009, April 1988.
Anderson, Brian D.O., ”A Note on Transmission Zeros of a Transfer Function Matrix,”
IEEE Trans. Automat. Contr., Vol. AC-21, No. 4, pp. 589-591, August 1976.

31
Antsaklis, Panos J., ”Maximal Order Reduction and Supremal (A,B)-Invariant and
Controllability Subspaces,” IEEE Trans. Automat. Contr., Vol. AC-25, No. 1, pp. 44-
49, Feb. 1980; ”On Finite and Infinite Zeros in the Model Matching Problem,” Proc.
25th Conf. Decision Contr., pp. 1295-1299, Dec. 1986.
Antsaklis, P.J., and Sain, M.K., ”Unity Feedback Compensation of Unstable Plants,”
Proc. 20th Conf. Decision Contr., pp. 305-308, Dec. 1981.
Antsaklis, P.J., and Williams, T.W.C., ”On the Dimensions of the Supremal (A,B)-
Invariant and Controllability Subspaces,” IEEE Trans. Automat. Contr., Vol. AC-25,
No. 6, pp. 1223-1225, Dec. 1980.
Araki, Mituhiko, ”A Note on the Calculation and Placing of Finite Zeros,” Int. J.
Contr., Vol. 37, No. 4, pp. 873-878, April 1983.
Åström, K.J., Hagander, P., and Sternby, J., ”Zeros of Sampled Systems,” Automatica,
Vol. 20, No. 1, pp. 31-38, Jan. 1984.
Athanassiades, M., Falb, P., Kalman, R.E., and Lee, E.B., ”Time Optimal Control for
Plants with Numerator Dynamics,” IRE Trans. Automat. Contr., Vol. AC-7, No. 4,
pp. 47-50, July 1962.
Axelby, G.S., and Davison, E.J., ”On the Computation of Transmission Zeros of Linear
Multivariable Systems,” Automatica, Vol. 12, No. 5, pp. 533-534, Sept. 1976.
Axelby, G.S., Laub, A.J., and Davison, E.J., ”Further Discussion on the Calculation of
Transmission Zeros,” Automatica, Vol. 14, No. 4, pp. 403-405, July 1978.
Ball, Joseph A., Cohen, Nir, and Rodman, Leiba, ”Zero Data and Interpolation Prob-
lems for Rectangular Matrix Polynomials,” Department of Mathematics, Virginia Poly-
technic Institute & State University, Blacksburg, VA, 1989.
Ball, Joseph A., Gohberg, Israel, and Rodman, Leiba, ”Tangential Interpolation Prob-
lems for Rational Matrix Functions,” Department of Mathematics, Virginia Polytechnic
Institute & State University, Blacksburg, VA, 1989.
Barnard, Robert D., ”An L-infinity Extension of Transmission-zero and Robust-design
Notions to Global Tracking in Uncertain Nonlinear Systems,” Proc. 27th Conf. Decision
Contr., pp. 482-485, Dec. 1988.
Bengtsson, Gunnar, ”Minimal System Inverses for Linear Multivariable Systems,” J.

32
Math. Anal. and Appl., Vol. 46, No. 1, pp. 261-274, April 1974.
Bollinger, K.E., and Mathur, J.C., ”To Compute the Zeros of Large Systems,” IEEE
Trans. Automat. Contr., Vol. AC-16, No. 1, pp. 95-96, Feb. 1971.
Bosgra, O.H., and Van Der Weiden, A.J.J., ”Realizations in Generalized State-space
Form for Polynomial System Matrices and the Definitions of Poles, Zeros and Decou-
pling Zeros at Infinity,” Int. J. Contr., Vol. 33, No. 3, pp. 393-411, March 1981.
Boullion, Thomas L., and Odell, Patrick L., Generalized Inverse Matrices. New York:
John Wiley & Sons, Inc., 1971.
Bristol, E.H., ”The Right Half Plane’ll Get You if You Don’t Watch Out,” Proc. 1981
Joint Automat. Contr. Conf., Vol. 1, Paper TA-7A, June 1981.
Brockett, Roger W., ”Poles, Zeros, and Feedback: State Space Interpretation,” IEEE
Trans. Automat. Contr., Vol. AC-10, No. 2, pp. 129-135, April 1965.
Byrnes, Christopher I., and Isidori, Alberto, ”A Frequency Domain Philosophy for
Nonlinear Systems, with Applications to Stabilization and to Adaptive Control,” Proc.
23rd Conf. Decision Contr., pp. 1569-1573, Dec. 1984; ”Global Feedback Stabilization
of Nonlinear Systems,” Proc. 24th Conf. Decision Contr., pp. 1031-1037, Dec. 1985;
”Feedback Design from the Zero Dynamics Point of View,” Dipartimento di Informatica
e Sistemistica, Università di Roma ”La Sapienza”, Roma, Italy, April 1988a; ”Local
Stabilization of Minimum-phase Nonlinear Systems,” Systems & Contr. Letters, Vol.
11, No. 1, July 1988b.
Callier, Frank M., Cheng, Victor Hon Lam, and Desoer, Charles A., ”Dynamic Interpre-
tation of Poles and Transmission Zeros for Distributed Multivariable Systems,” IEEE
Trans. Circuits Syst., Vol. CAS-28, No. 4, pp. 300-306, April 1981.
Cameron, R.G., and Zolghadri-Jahromi, M., ”Zero Assignment by Dyadic Output Feed-
back,” Int. J. Contr., Vol. 36, No. 3, pp. 419-427, Sept. 1982.
Chang, B.-C., and Banda, S.S., ”Optimal H ∞ Norm Computation for Multivariable
Systems with Multiple Zeros,” IEEE Trans. Automat. Contr., Vol. 34, No. 5, pp. 553-
557, May 1989.
Cheng, V.H.L., and Desoer, C.A., ”Limitations on the Closed-Loop Transfer Function
Due to Right-Half Plane Transmission Zeros of the Plant,” IEEE Trans. Automat.

33
Contr., Vol. AC-25, No. 6, pp. 1218-1220, Dec. 1980.
Clarke, D.W., ”Self-tuning Control of Nonminimum-phase Systems,” Automatica, Vol.
20, No. 5, pp. 501-517, Sept. 1984.
Commault, C., and Dion, J.M., ”Structure at Infinity of Linear Multivariable Systems:
A Geometric Approach,” IEEE Trans. Automat. Contr., Vol. AC-27, No. 3, pp. 693-
696, June 1982.
Conte, G., Longhi, S., and Perdon, A., ”Structure at Infinity for Discrete Time Linear
Periodic Systems,” Proc. 27th Conf. Decision Contr., pp. 902-903, Dec. 1988.
Conte, G., and Perdon, A.M., ”An Algebraic Notion of Zeros for Systems over Rings,”
Proc. Int. Symp. Math. Theory of Networks and Syst., July 1983; ”Infinite Zero Mod-
ule and Infinite Pole Module,” Proc. 6th Int. Conf. Anal. and Opt. of Syst. (reprint),
Lecture Notes Contr. and Inf. Sci., Vol. 62, 1984; ”Zero Modules and Factorization
Problems,” Contemporary Mathematics: Linear Algebra and Its Role in Systems The-
ory. Providence: American Mathematical Society, Vol. 47, pp. 81-94, 1985a; ”On the
Causal Factorization Problem,” IEEE Trans. Automat. Contr., Vol. AC-30, No. 8, pp.
811-813, August 1985b; ”Zeros of Cascade Compositions,” Frequency Domain and State
Space Methods for Linear Systems, C.I. Byrnes and A. Lindquist, eds., North Holland:
Elsevier Science Publishers B. V., 1986.
Conte, G., Perdon, A.M., and Wyman, B.F., ”Factorizations of Transfer Functions,”
Proc. 25th Conf. Decision Contr., pp. 1279-1283, Dec. 1986; ”Fixed Poles in Transfer
Function Equations,” SIAM J. Contr. and Opt., Vol. 26, No. 2, pp. 356-368, March
1988.
Corfmat, J.P., and Morse, A.S., ”Control of Linear Systems Through Specified Input
Channels,” SIAM J. Contr. and Opt., Vol. 14, No. 1, pp. 163-175, Jan. 1976.
D’Andrea, B., and Praly, L., ”About Finite Nonlinear Zeros for Decouplable Systems,”
Systems & Contr. Letters, Vol. 10, No. 2, pp. 103-109, Feb. 1988.
Davison, E.J., ”A Nonminimum Phase Index and its Application to Interacting Mul-
tivariable Control Systems,” Automatica, Vol. 5, No. 6, pp. 791-799, Nov. 1969; ”A
Computational Method for Finding the Zeros of a Multivariable Linear Time Invariant
System,” Automatica, Vol. 6, No. 3, pp. 481-484, May 1970.

34
Davison, E.J., and Özgüner, Ü., ”Characterizations of Decentralized Fixed Modes for
Interconnected Systems,” Automatica, Vol. 19, No. 2, pp. 169-182, March 1983.
Davison, E.J., and Wang, S.-H., ”Properties and Calculation of Transmission Zeros of
Linear Multivariable Systems,” Automatica, Vol. 10, No. 6, pp. 643-658, Dec. 1974,
Correction Item, Vol. 13, No. 3, p. 327, May 1977; ”Remark on Multiple Transmission
Zeros of a System,” Automatica, Vol. 12, No. 2, p. 195, March 1976; ”An Algorithm
for the Calculation of Transmission Zeros of the System (C, A, B, D) Using High Gain
Output Feedback,” IEEE Trans. Automat. Contr., Vol. AC-23, No. 4, pp. 738-741, Au-
gust 1978; ”A Characterization of Decentralized Fixed Modes in Terms of Transmission
Zeros,” IEEE Trans. Automat. Contr., Vol. AC-30, No. 1, pp. 81-82, Jan. 1985.
Demmel, James Weldon, and Kågström, Bo, ”Computing Stable Eigendecompositions
of Matrix Pencils,” Linear Algebra and Its Appl., Vol. 88/89, pp. 139-186, 1987; ”Accu-
rate Solutions of Ill-posed Problems in Control Theory,” SIAM J. Matrix Anal. Appl.,
Vol. 9, No. 1, pp. 126-145, Jan. 1988.
Desoer, Charles A., and Schulman, Jerry D., ”Zeros and Poles of Matrix Transfer Func-
tions and Their Dynamical Interpretation,” IEEE Trans. Circuits Syst., Vol. CAS-21,
No. 1, pp. 3-8, Jan. 1974.
Di Benedetto, M.D., and Slotine, J.-J., ”A Note on Zeros of Nonlinear Systems: In-
ferring Sliding Surface Selection from the Zero-dynamics Algorithm,” Proc. 27th Conf.
Decision Contr., pp. 915-920, Dec. 1988.
Dion, J.M., and Commault, C., ”The Minimal Delay Decoupling Problem: Feedback
Implementation with Stability,” SIAM J. Contr. and Opt., Vol. 26, No. 1, pp. 66-82,
Jan. 1988.
El-Ghezawi, O.M.E., Zinober, A.S.I., Owens, D.H., and Billings, S.A., ”Computation
of the Zeros and Zero Directions of Linear Multivariable Systems,” Int. J. Contr., Vol.
36, No. 5, pp. 833-843, Nov. 1982.
Elliott, Howard, and Wolovich, William A., ”A Parameter Adaptive Control Structure
for Linear Multivariable Systems,” IEEE Trans. Automat. Contr., Vol. AC-27, No. 2,
pp. 340-352, April 1982.
Emami-Naeini, Abbas, and Rock, Steve M., ”On Asymptotic Behavior of Non-square

35
Linear Optimal Regulators,” Proc. 23rd Conf. Decision Contr., pp. 1762-1763, Dec.
1984.
Emami-Naeini, A., and Van Dooren, P., ”Computation of Zeros of Linear Multivariable
Systems,” Automatica, Vol. 18, No. 4, pp. 415-430, July 1982a; ”On Computation of
Transmission Zeros and Transfer Functions,” Proc. 21st Conf. Decision Contr., pp.
51-55, Dec. 1982b.
Ferreira, Pedro M.G., ”Infinite System Zeros,” Int. J. Contr., Vol. 32, No. 4, pp.
731-735, Oct. 1980; ”Invariance of Infinite Zeros Under Feedback,” Int. J. Contr., Vol.
35, No. 3, pp. 535-543, March 1982; ”Comments On ’On Blocking Zeros in Linear
Multivariable Systems’,” IEEE Trans. Automat. Contr., Vol. AC-31, No. 12, pp. 1175-
1176, Dec. 1986.
Ferreira, P.G., and Bhattacharyya, S.P., ”On Blocking Zeros,” IEEE Trans. Automat.
Contr., Vol. AC-22, No. 2, pp. 258-259, April 1977.
Francis, B.A., and Wonham, W.M., ”The Role of Transmission Zeros in Linear Multi-
variable Regulators,” Int. J. Contr., Vol. 22, No. 5, pp. 657-681, Nov. 1975.
Franklin, G.F., and Johnson, C. Richard, Jr., ”A Condition for Full Zero Assignment
in Linear Control Systems,” IEEE Trans. Automat. Contr., Vol. AC-26, No. 2, pp.
519-521, April 1981.
Freudenberg, James S., and Looze, Douglas P., ”Right Half Plane Poles and Zeros and
Design Tradeoffs in Feedback Systems,” IEEE Trans. Automat. Contr., Vol. AC-30, No.
6, pp. 555-565, June 1985.
Fu, Ye, and Dumont, Guy A., ”Choice of Sampling to Ensure Minimum-Phase Behav-
ior,” IEEE Trans. Automat. Contr., Vol. 34, No. 5, pp. 560-563, May 1989.
Glover, K., and Silverman, L.M., ”Characterization of Structural Controllability,” IEEE
Trans. Automat. Contr., Vol. AC-21, No. 4, pp. 534-537, August 1976.
Grizzle, J.W., and Shor, M.H., ”Sampling, Infinite Zeros and Decoupling of Linear
Systems,” Automatica, Vol. 24, No. 3, pp. 387-396, May 1988.
Guidorzi, Roberto P., ”On the Algorithms for Computing the Zeros of Large Systems,”
IEEE Trans. Automat. Contr., Vol. AC-17, No. 5, pp. 731-732, Oct. 1972.
Guidorzi, R., and Terragni, F., ”Zeros Determination in Large-scale Multivariable Sys-

36
tems,” Int. J. Contr., Vol. 17, No. 4, pp. 747-752, April 1973.
Guillemin, Ernst A., Synthesis of Passive Networks. New York: John Wiley & Sons,
Inc., 1957.
Harvey, Charles A., and Stein, Gunter, ”Quadratic Weights for Asymptotic Regulator
Properties,” IEEE Trans. Automat. Contr., Vol. AC-23, No. 3, pp. 378-387, June 1978.
Hewer, G.A., and Martin, J.M., ”An Exact Characterization of Linear Systems with
Zeros Everywhere in the Complex Plane,” IEEE Trans. Automat. Contr., Vol. AC-29,
No. 8, pp. 728-730, August 1984.
Horan, R.E., ”On the Decoupling Zeros and Poles of a System,” IEEE Trans. Automat.
Contr., Vol. AC-25, No. 3, pp. 517-521, June 1980.
Hosoe, S., ”On a Time-Domain Characterization of the Numerator Polynomials of the
Smith McMillan form,” IEEE Trans. Automat. Contr., Vol. AC-20, No. 6, pp. 799-800,
Dec. 1975.
Hung, Y.S., and MacFarlane, A.G.J., ”On the Relationships Between the Unbounded
Asymptote Behaviour of Multivariable Root Loci, Impulse Response and Infinite Zeros,”
Int. J. Contr., Vol. 34, No. 1, pp. 31-69, July 1981.
Isidori, A., and Moog, C., ”On the Nonlinear Equivalent of the Notion of Transmission
Zeros,” Modeling and Adaptive Control; Lect. Notes Contr. and Inf. Sci., C.I. Byrnes
and A.H. Kurszanski, eds., Springer-Verlag, Vol. 105, 1988.
Kailath, Thomas, Linear Systems. New Jersey: Prentice-Hall, Inc., 1980.
Kalnitsky, K.C., and Kwatny, H.G., ”An Eigenvalue Characterization of Multivariable
System Zeros,” IEEE Trans. Automat. Contr., Vol. AC-22, No. 2, pp. 259-262, April
1977.
Kamen, Edward W., ”The Poles and Zeros of a Linear Time-Varying System,” Linear
Algebra and Its Appl., Vol. 98, pp. 263-289, 1988a; ”On the Inner and Outer Poles
and Zeros of a Linear Time-Varying System,” Proc. 27th Conf. Decision Contr., pp.
910-914, Dec. 1988b.
Karcanias, N., and Kouvaritakis, B., ”The Output Zeroing Problem and Its Relationship
to the Invariant Zero Structure: a Matrix Pencil Approach,” Int. J. Contr., Vol. 30,
No. 3, pp. 395-415, Sept. 1979.

37
Karcanias, N., Laios, B., and Giannakopoulos, C., ”Decentralized Determinantal As-
signment Problem: Fixed and Almost Fixed Modes and Zeros,” Int. J. Contr., Vol. 48,
No. 1, pp. 129-147, July 1988.
Kaufman, Ilia, ”On Poles and Zeros of Linear Systems,” IEEE Trans. Circuit Theory,
Vol. CT-20, No. 2, pp. 93-101, March 1973.
Kimura, Hidenori, ”Perfect and Subperfect Regulation in Linear Multivariable Control
Systems,” Automatica, Vol. 18, No. 2, pp. 125-145, March 1982.
Kouvaritakis, B., ”A Geometric Approach to the Inversion of Multivariable Systems,”
Int. J. Contr., Vol. 24, No. 5, pp. 609-626, Nov. 1976.
Kouvaritakis, B., and Daniel, R.W., ”A Comparison Between the NAM and Related
Techniques with a Recent Method for the Computation of Zeros and Zero Directions,”
Int. J. Contr., Vol. 39, No. 3, pp. 581-586, March 1984.
Kouvaritakis, B., and Edmunds, J.M., ”Multivariable Root Loci: a Unified Approach
to Finite and Infinite Zeros,” Int. J. Contr., Vol. 29, No. 3, pp. 393-428, March 1979.
Kouvaritakis, B., and MacFarlane, A.G.J., ”Geometric Approach to Analysis and Syn-
thesis of System Zeros: Part 1. Square Systems,” Int. J. Contr., Vol. 23, No. 2, pp.
149-166, Feb. 1976a; ”Geometric Approach to Analysis and Synthesis of System Zeros:
Part 2. Non-square Systems,” Int. J. Contr., Vol. 23, No. 2, pp. 167-181, Feb. 1976b.
Kouvaritakis, B., and Shaked, U., ”Asymptotic Behaviour of Root-loci of Linear Mul-
tivariable Systems,” Int. J. Contr., Vol. 23, No. 3, pp. 297-340, March 1976.
Krener, A.J., and Isidori, A., ”Nonlinear Zero Distributions,” Proc. 19th Conf. Decision
Contr., pp. 665-668, Dec. 1980.
Kwatny, H.G., and Bennett, W.H., ”Nonlinear Dynamics and Control Issues for Flexible
Space Platforms,” Proc. 27th Conf. Decision Contr., pp. 1702-1707, Dec. 1988.
Laub, A.J., and Moore, B.C., ”Calculation of Transmission Zeros Using QZ Tech-
niques,” Automatica, Vol. 14, No. 6, pp. 557-566, Nov. 1978.
Lee, Soodong, Meerkov, Semyon M., and Runolfsson, Thordur, ”Vibrational Feedback
Control: Zeros Placement Capabilities,” IEEE Trans. Automat. Contr., Vol. AC-32,
No. 7, pp. 604-611, July 1987.
Lewis, F.L., ”A Survey of Linear Singular Systems,” Circuits, Systems, Signal Process.,

38
Vol. 5, No. 1, pp. 3-36, 1986.
Lindner, Douglas Kent, ”Zeros of Multivariable Systems: Definitions and Algorithms,”
M.S. Thesis, University of Illinois, Urbana, IL, May 1979.
Lovass-Nagy, V., and Powers, D.L., ”Determination of Zeros and Zero Directions of
Linear Time-invariant Systems by Matrix Generalized Inverses,” Int. J. Contr., Vol.
31, No. 6, pp. 1161-1170, June 1980.
MacFarlane, A.G.J., ”Relationships Between Recent Developments in Linear Control
Theory and Classical Design Techniques,” Measurement and Control, Vol. 8, May-Sept.
1975.
MacFarlane, A.G.J., and Karcanias, N., ”Poles and Zeros of Linear Multivariable Sys-
tems: a Survey of the Algebraic, Geometric and Complex-Variable Theory,” Int. J.
Contr., Vol. 24, No. 1, pp. 33-74, July 1976.
MacFarlane, A.G.J., Kouvaritakis, B., and Edmunds, J.M., ”Complex Variable Meth-
ods for Multivariable Feedback Systems Analysis and Design,” Alternatives for Linear
Multivariable Control, Michael K. Sain, Joseph L. Peczkowski, and James L. Melsa,
eds., National Engineering Consortium, Inc., pp. 189-228, 1978.
MacFarlane, A.G.J., and Postlethwaite, I., ”The Generalized Nyquist Stability Criterion
and Multivariable Root Loci,” Int. J. Contr., Vol. 25, No. 1, pp. 81-127, Jan. 1977.
Maione, Bruno, and Turchiano, Biagio, ”Characterization of Decoupling Structural Ze-
ros of Rosenbrock’s Polynomial Matrices,” Int. J. Contr., Vol. 47, No. 2, pp. 459-476,
Feb. 1988.
Malabre, M., and Kučera, V., ”Infinite Structure and Exact Model Matching Problem:
A Geometric Approach,” IEEE Trans. Automat. Contr., Vol. AC-29, No. 3, pp. 266-268,
March 1984.
Marshall, S.A., ”Remarks on Computing the Zeros of Large Systems,” IEEE Trans.
Automat. Contr., Vol. AC-17, No. 2, p. 261, April 1972.
Misra, P., and Patel, R.V., ”Transmission Zero Assignment in Linear Multivariable
Systems Part I: Square Systems,” Proc. 27th Conf. Decision Contr., pp. 1310-1311,
Dec. 1988.
Mita, T., ”On Maximal Unobservable Subspace, Zeros and Their Applications,” Int. J.

39
Contr., Vol. 25, No. 6, pp. 885-899, June 1977.
Moler, C.B., and Stewart, G.W., ”An Algorithm for Generalized Matrix Eigenvalue
Problems,” SIAM J. Numer. Anal., Vol. 10, pp. 241-256, 1973.
Molinari, B.P., ”Zeros of the System Matrix,” IEEE Trans. Automat. Contr., Vol. AC-
21, No. 5, pp. 795-797, Oct. 1976.
Moog, Claude H., ”Nonlinear Decoupling and Structure at Infinity,” Mathematics of
Contr., Signals and Syst., Vol. 1, pp. 257-268, 1988.
Moore, Bruce C., and Laub, Alan J., ”Computation of Supremal (A,B)-Invariant and
Controllability Subspaces,” IEEE Trans. Automat. Contr., Vol. AC-23, No. 5, pp. 783-
792, Oct. 1978; ”Correction to ’Computation of Supremal (A,B)-Invariant and Control-
lability Subspaces’,” IEEE Trans. Automat. Contr., Vol. AC-24, No. 4, p. 677, August
1979.
Moore, B.C., and Silverman, L.M., ”A Time Domain Characterization of the Invariant
Factors of a System Transfer Function,” Proc. 1974 Joint Automat. Contr. Conf., pp.
186-193, June 1974.
Morse, A.S., ”Structural Invariants of Linear Multivariable Systems,” SIAM J. Contr.,
Vol. 11, No. 3, pp. 446-465, August 1973; Private Communication, Feb. 1989.
M’Saad, M., Ortega, R., and Landau, I.D., ”Adaptive Controllers for Discrete-Time
Systems with Arbitrary Zeros: An Overview,” Automatica, Vol. 21, No. 4, pp. 413-423,
July 1985.
Murdoch, P., ”Pole and Zero Assignment by Proportional Feedback,” IEEE Trans.
Automat. Contr., Vol. AC-18, No. 5, p. 542, Oct. 1973.
Nijmeijer, Henk, and Schumacher, Johannes M., ”Zeros at Infinity for Affine Nonlinear
Control Systems,” IEEE Trans. Automat. Contr., Vol. AC-30, No. 6, pp. 566-573, June
1985.
Ohkami, Y., and Likins, P.W., ”Determination of Poles and Zeros of Transfer Functions
for Flexible Spacecraft Attitude Control,” Int. J. Contr., Vol. 24, No. 1, pp. 13-22, July
1976.
Owens, D.H., ”Invariant Zeros of Multivariable Systems: A Geometric Analysis,” Int.
J. Contr., Vol. 26, No. 4, pp. 537-548, Oct. 1977.

40
O’Young, S.D., Postlethwaite, I., and Gu, D.-W., ”A Treatment of jω-Axis Model-
Matching Transformation Zeros in the Optimal H 2 and H ∞ Control Designs,” IEEE
Trans. Automat. Contr., Vol. 34, No. 5, pp. 551-553, May 1989.
Özgüler, A. Bülent, and Eldem, Vasfï, ”Disturbance Decoupling Problems Via Dynamic
Output Feedback,” IEEE Trans. Automat. Contr., Vol. AC-30, No. 8, pp. 756-764,
August 1985.
Pandolfi, Luciano, ”The Transmission Zeros of Systems with Delays,” Int. J. Contr.,
Vol. 36, No. 6, pp. 959-976, Dec. 1982.
Passino, K.M., and Antsaklis, P.J., ”Inverse Stable Sampled Low-pass Systems,” Int.
J. Contr., Vol. 47, No. 6, pp. 1905-1913, June 1988.
Patel, R.V., ”On the Computation of Numerators of Transfer Functions of Linear Sys-
tems,” IEEE Trans. Automat. Contr., Vol. AC-18, No. 4, pp. 400-401, August 1973;
”On Zeros of Multivariable Systems,” Int. J. Contr., Vol. 21, No. 4, pp. 599-608, April
1975; ”On Computing the Invariant Zeros of Multivariable Systems,” Int. J. Contr.,
Vol. 24, No. 1, pp. 145-146, July 1976; ”On Transmission Zeros and Dynamic Output
Feedback,” IEEE Trans. Automat. Contr., Vol. AC-23, No. 4, pp. 741-742, August 1978;
”On Blocking Zeros in Linear Multivariable Systems,” IEEE Trans. Automat. Contr.,
Vol. AC-31, No. 3, pp. 239-241, March 1986.
Patel, R.V., Sinswat, V., and Fallside, F., ”’Disturbance Zeros’ in Multivariable Sys-
tems,” Int. J. Contr., Vol. 26, No. 1, pp. 85-96, July 1977.
Pohjolainen, Seppo, ”Computation of Transmission Zeros for Distributed Parameter
Systems,” Int. J. Contr., Vol. 33, No. 2, pp. 199-212, Feb. 1981.
Porter, B., ”Invariant Zeros and Zero-directions of Multivariable Linear Systems with
Slow and Fast Modes,” Int. J. Contr., Vol. 28, No. 1, pp. 81-91, July 1978.
Porter, B., and Jones, A.H., ”Time-Domain Identification of Transmission Zero Loca-
tions of Linear Multivariable Plants,” IEEE Trans. Automat. Contr., Vol. AC-30, No.
10, pp. 1050-1053, Oct. 1985.
Postlethwaite, Ian, and Foo, Yung Kuan, ”Robustness with Simultaneous Pole and Zero
Movement across the jω-Axis,” Automatica, Vol. 21, No. 4, pp. 433-443, July 1985.
Pugh, A.C., ”Transmission and System Zeros,” Int. J. Contr., Vol. 26, No. 2, pp. 315-

41
324, August 1977.
Pugh, A.C., and Ratcliffe, P.A., ”On the Zeros and Poles of a Rational Matrix,” Int. J.
Contr., Vol. 30, No. 2, pp. 213-226, August 1979.
Rosenbrock, H.H., State-space and Multivariable Theory. New York: John Wiley &
Sons, Inc., 1970; ”The Zeros of a System,” Int. J. Contr., Vol. 18, No. 2, pp. 297-299,
August 1973; ”Structural Properties of Linear Dynamical Systems,” Int. J. Contr., Vol.
20, No. 2, pp. 191-202, August 1974a; ”Correction to ’The Zeros of a System’,” Int.
J. Contr., Vol. 20, No. 3, pp. 525-527, Sept. 1974b; ”Comments on ’Poles and Zeros
of Linear Multivariable Systems: a Survey of the Algebraic, Geometric and Complex-
Variable Theory’,” Int. J. Contr., Vol. 26, No. 1, pp. 157-161, July 1977.
Rosenbrock, H.H., and Pugh, A.C., ”Contributions to a Hierarchical Theory of Sys-
tems,” Int. J. Contr., Vol. 19, No. 5, pp. 845-867, May 1974.
Rosenbrock, H.H., and Van Der Weiden, A.J.J., ”Inverse Systems,” Int. J. Contr., Vol.
25, No. 3, pp. 389-392, March 1977.
Sain, Michael K., Wyman, Bostwick F., Gejji, R.R., Antsaklis, P.J., and Peczkowski,
Joseph L., ”The Total Synthesis Problem of Linear Multivariable Control Part I: Nom-
inal Design,” Proc. 1981 Joint Automat. Contr. Conf., Vol. 1, Paper WP-4A, June
1981a; ”The Total Synthesis Problem of Linear Multivariable Control Part II: Unity
Feedback and the Design Morphism,” Proc. 20th Conf. Decision Contr., pp. 875-884,
Dec. 1981b.
Sain, Michael K., Wyman, Bostwick F., and Peczkowski, Joseph L., ”Matching Zeros:
a Fixed Constraint in Multivariable Synthesis,” Proc. 27th Conf. Decision Contr., pp.
2060-2065, Dec. 1988.
Sannuti, P., and Saberi, A., ”Special Coordinate Basis for Multivariable Linear Systems—
Finite and Infinite Zero Structure, Squaring Down and Decoupling,” Int. J. Contr., Vol.
45, No. 5, pp. 1655-1704, May 1987.
Schrader, Cheryl B., and Sain, Michael K., ”Subzeros in Feedback Transmission,” Proc.
1989 American Contr. Conf., June 1989.
Seraji, H., ”Design of Cascade Controllers for Zero Assignment in Multivariable Sys-
tems,” Int. J. Contr., Vol. 21, No. 3, pp. 485-496, March 1975.

42
Shaked, U., ”The Zero Properties of Linear Passive Systems,” IEEE Trans. Automat.
Contr., Vol. AC-22, No. 6, pp. 973-974, Dec. 1977.
Shaked, U., and Karcanias, N., ”The Use of Zeros and Zero-directions in Model Reduc-
tion,” Int. J. Contr., Vol. 23, No. 1, pp. 113-135, Jan. 1976.
Shaked, U., and Kouvaritakis, B., ”The Zeros of Linear Optimal Control Systems and
Their Role in High Feedback Gain Stability Design,” IEEE Trans. Automat. Contr.,
Vol. AC-22, No. 4, pp. 597-599, August 1977.
Silverman, L.M., and Van Dooren, P., ”A System Theoretic Approach for GCD Extrac-
tion,” Proc. 17th Conf. Decision Contr., pp. 525-528, Jan. 1979.
Simon, J.D., and Mitter, S.K., ”Synthesis of Transfer Function Matrices with Invariant
Zeros,” IEEE Trans. Automat. Contr., Vol. AC-14, No. 4, pp. 420-421, August 1969.
Sinswat, V., and Fallside, F., ”On Minimal-order Inverses of Linear Continuous Time-
invariant Systems,” Int. J. Contr., Vol. 24, No. 6, pp. 853-868, Dec. 1976; ”Determi-
nation of Invariant Zeros and Transmission Zeros of All Classes of Invertible Systems,”
Int. J. Contr., Vol. 26, No. 1, pp. 97-114, July 1977.
Sinswat, V., Patel, R.V., and Fallside, F., ”A Method for Computing Invariant Zeros
and Transmission Zeros of Invertible Systems,” Int. J. Contr., Vol. 23, No. 2, pp. 183-
196, Feb. 1976.
Smith, M.C., ”On the Generalized Nyquist Stability Criterion,” Int. J. Contr., Vol. 34,
No. 5, pp. 885-920, Nov. 1981; ”Applications of Algebraic Function Theory in Multivari-
able Control,” Multivariable Control: New Concepts and Tools, S.G. Tzafestas, ed., D.
Reidel Publishing Co., 1984; ”Multivariable Root-locus Behaviour and the Relationship
to Transfer-function Pole-zero Structure,” Int. J. Contr., Vol. 43, No. 2, pp. 497-515,
Feb. 1986.
Smithberger, Gregory, and Wyman, Bostwick F., ”Zeros and Periodicity for Systems
over Finite Fields,” Proc. 27th Conf. Decision Contr., pp. 904-909, Dec. 1988.
Suda, N., and Mutsuyoshi, E., ”Invariant Zeros and Input-output Structure of Linear,
Time-invariant Systems,” Int. J. Contr., Vol. 28, No. 4, pp. 525-535, Oct. 1978.
Tarokh, M., ”Fixed Modes in Multivariable Systems using Constrained Controllers,”
Automatica, Vol. 21, No. 4, pp. 495-497, July 1985.

43
Tsai, Te-Ping, and Wang, Te-Shing, ”Optimal Design of Non-minimum-phase Control
Systems with Large Plant Uncertainty,” Int. J. Contr., Vol. 45, No. 6, pp. 2147-2159,
June 1987.
Vandewalle, J., and Dewilde, P., ”On the Determination of the Order and the Degree
of a Zero of a Rational Matrix,” IEEE Trans. Automat. Contr., Vol. AC-19, No. 5, pp.
608-609, Oct. 1974.
Van Dooren, P., Emami-Naeini, A., and Silverman, L., ”Stable Extraction of the Kro-
necker Structure of Pencils,” Proc. 17th Conf. Decision Contr., pp. 521-524, Jan. 1979.
Vardulakis, Antonis I.G., ”Zero Placement and the ’Squaring Down’ Problem: a Poly-
nomial Matrix Approach,” Int. J. Contr., Vol. 31, No. 5, pp. 821-832, May 1980a; ”On
Infinite Zeros,” Int. J. Contr., Vol. 32, No. 5, pp. 849-866, Nov. 1980b.
Vardulakis, A.I.G., Limebeer, D.N.J, and Karcanias, N., ”Structure and Smith-MacMillan
Form of a Rational Matrix at Infinity,” Int. J. Contr., Vol. 35, No. 4, pp. 701-725, April
1982.
Verghese, G., Van Dooren, P., and Kailath, T., ”Properties of the System Matrix of a
Generalized State-space System,” Int. J. Contr., Vol. 30, No. 2, pp. 235-243, August
1979.
Vidyasagar, M., ”On Undershoot and Nonminimum Phase Zeros,” IEEE Trans. Au-
tomat. Contr., Vol. AC-31, No. 5, p. 440, May 1986.
Williams, Trevor, ”Computing the Transmission Zeros of Large Space Structures,” IEEE
Trans. Automat. Contr., Vol. 34, No. 1, pp. 92-94, Jan. 1989a; ”Transmission Zero
Bounds for Large Space Structures, with Applications,” AIAA J. Guidance, Contr.,
and Dynamics, Vol. 12, pp. 33-38, Jan.-Feb. 1989b.
Williams, T.W.C., and Antsaklis, P.J., ”A Unifying Approach to the Decoupling of
Linear Multivariable Systems,” Int. J. Contr., Vol. 44, No. 1, pp. 181-201, July 1986.
Williams, Trevor, and Juang, Jer-Nan, ”Pole/Zero Cancellations in Flexible Space
Structures,” Proc. AIAA Guidance, Navigation and Contr. Conf., Vol. 1, pp. 33-40,
Aug. 1988.
Wolovich, W.A., ”On Determining the Zeros of State-Space Systems,” IEEE Trans.
Automat. Contr., Vol. AC-18, No. 5, pp. 542-544, Oct. 1973a; ”On the Numerators

44
and Zeros of Rational Transfer Matrices,” IEEE Trans. Automat. Contr., Vol. AC-18,
No. 5, pp. 544-546, Oct. 1973b; ”On the Cancellation of Multivariable System Zeros by
State Feedback,” IEEE Trans. Automat. Contr., Vol. AC-19, No. 3, pp. 276-277, June
1974.
Wolovich, W.A., Antsaklis, P., and Elliott, H., ”On the Stability of Solutions to Minimal
and Nonminimal Design Problems,” IEEE Trans. Automat. Contr., Vol. AC-22, No. 1,
pp. 88-94, Feb. 1977.
Wolovich, W.A., and Falb, P.L., ”Invariants and Canonical Forms under Dynamic Com-
pensation,” SIAM J. Contr. and Opt., Vol. 14, No. 6, pp. 996-1008, Nov. 1976.
Wonham, W. Murray, Linear Multivariable Control; A Geometric Approach. New York:
Springer-Verlag, 1974.
Wyman, Bostwick F., and Sain, Michael K., ”The Zero Module and Invariant Sub-
spaces,” Proc. 19th Conf. Decision Contr., pp. 254-255, Dec. 1980; ”The Zero Mod-
ule and Essential Inverse Systems,” IEEE Trans. Circuits Syst., Vol. CAS-28, No. 2,
pp. 112-126, Feb. 1981a; ”The Pole Structure of Inverse Systems,” Int. Fed. Automat.
Contr.; 8th Triennial World Cong. (preprints), Vol. 3, pp. 76-81, August 1981b; ”Exact
Sequences for Pole-Zero Cancellation,” Proc. Int. Symp. Math. Theory of Networks and
Syst., pp. 278-280, August 1981c; ”Internal Zeros and the System Matrix,” Proc. 20th
Allerton Conf. on Communication, Contr., and Computing, pp. 153-158, Oct. 1982;
”On the Zeros of a Minimal Realization,” Linear Algebra and Its Appl., Vol. 50, pp.
621-637, 1983; ”On the Design of Pole Modules for Inverse Systems,” IEEE Trans. Cir-
cuits Syst., Vol. CAS-32, No. 10, pp. 977-988, Oct. 1985; ”On Dual Zero Spaces and
Inverse Systems,” IEEE Trans. Automat. Contr., Vol. AC-31, No. 11, pp. 1053-1055,
Nov. 1986; ”Module Theoretic Zero Structures for System Matrices,” SIAM J. Contr.
and Opt., Vol. 25, No. 1, pp. 86-99, Jan. 1987.
Wyman, Bostwick F., Sain, Michael K., Conte, Giuseppe, and Perdon, Anna-Maria,
”Rational Matrices: Counting the Poles and Zeros,” Proc. 27th Conf. Decision Contr.,
pp. 921-925, Dec. 1988; ”On the Zeros and Poles of a Transfer Function,” Linear Algebra
and Its Appl., 1989.
Young, Kar-Keung D., Kokotović, Petar V., and Utkin, Vadim I., ”A Singular Pertur-

45
bation Analysis of High-Gain Feedback Systems,” IEEE Trans. Automat. Contr., Vol.
AC-22, No. 6, pp. 931-938, Dec. 1977.
Zwart, Hans, ”Characterization of All Controlled Invariant Subspaces for Spectral Sys-
tems,” SIAM J. Contr. and Opt., Vol. 26, No. 2, pp. 369-386, March 1988.

46

You might also like